自动驾驶中点云与图像多模态融合研究综述

Research Review of Multimodal Fusion of Point Cloud and Image in Autonomous Driv-ing

针对复杂多变的道路环境,综合国内外研究现状,本文从激光雷达和摄像头方面论述了汽车自动驾驶中的网络输入的格式,并以两种传感器融合为例,归纳了自动驾驶汽车环境感知任务中多模态传感器融合的分类方法,在此基础上,又从融合阶段的角度总结出另一种分类,简化了融合方法的分类和理解,强调了融合程度的区别以及融合方法的整体性,这种分类对于推动融合方法的研究和发展具有创新价值。最后分析传感器融合所遗留的问题,对未来的发展趋势进行预测。

In view of the complex and changeable road environment, this paper discusses the format of net-work input in auto driving from the aspects of laser radar and camera, and summarizes the classification method of multimodal sensor fusion in the environmental perception task of autonomous vehicle, based on which, another classification is summarized from the perspective of fusion stage, simplifying the classification and understanding of fusion methods, emphasizing the differences in fusion levels and the integrity of fusion methods, and this classification has innovative value for promoting the research and development of fusion methods. Finally, the issues left by sensor fusion and predict future development trends is analyzed.